Section: Application Domains

Crowd-sourced content recommendation

The Internet today serves as a large content distribution platform (online content varies from traditional news, TV series, and movies to specialized blogs and family pictures shared over social networks) as well as a platform for users to exchange opinions about practically everything (from movies to services and restaurants). The amount of information available online today overwhelms most users and selecting which content to watch or what do has become a challenge. We are applying passive measurement methods and content summarisation techniques to help users to identify relevant content in two scenarios. First, we are developing a system called WeBrowse that passively observes network traffic to extract user clicks (i.e., the URLs users visit). A user click is a good measure of interest, as users often have an idea of the type of content they are about to access (e.g., because they saw a preview or because a friend recommended it). Intuitively, the more users click on a URL, the higher the interest in the content on the corresponding page. WeBrowse then promotes “hottest” and most popular content to users of a network. We have a deployment of WeBrowse in a campus network. Second, we are working on techniques to summarise user feedback (for example, movie or restaurant reviews) with semi-structured feedback. Today reviews are either free-form text or star rating. Star rating is too coarse to capture the nuances of why a user likes or dislikes something, whereas free text is hard for users to parse and extract a clear opinion. We are instead working with semi-structured reviewing where users enter tags (a short sequence of words describing the user experience). We are working with Technicolor on the summarisation of movie reviews and on building a mobile app (called TagIt) where users can review movies directly with tags.